Neural networks and other machine-learning systems are used to create automatic financial forecasting and trading systems. To aid comparison of such systems, there is a need for reliable performance metrics. One such metric that may be considered is the win rate. We show how in certain circumstances the win-rate statistic can be very misleading, and to counter this, we propose and define baseline win rates for comparison. We develop empirical and closed-form models for such baselines and validate them against financial data and a neural forecaster.
|Publication status||Published - 19 Jul 2020|
|Event||IEEE WCCI 2020 - Glasgow|
Duration: 19 Jul 2020 → …
|Conference||IEEE WCCI 2020|
|Period||19/07/20 → …|